Research map / 3D vision / Face and shape
Diffusion-Based Material Regularization for Physics-Based Inverse Rendering
Jingwang Ling, Lifan Wu, Feng Xu, Shuang Zhao
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add implicit material clustering regularization (ablation) | PBIR pipeline · Synthetic4Relight | relighting PSNR: 28.94 -> 32.02 | Improved |
| add implicit material clustering regularization (ablation) | PBIR pipeline · DTC-Synthetic | relighting PSNR: 40.23 -> 43.21 | Improved |
| add implicit material clustering regularization (ablation) | PBIR pipeline · Stanford-ORB relighting | PSNR-H: 26.11 -> 27.22 | Improved |
| add regularization to Synthetic4Relight albedo estimate (ablation) | PBIR pipeline · Synthetic4Relight albedo | PSNR: 23.79 -> 27.04 | Improved |
| use clustering regularizer instead of global scale-invariant loss (ablation) | PBIR pipeline · Stanford-ORB relighting | PSNR-H: 26.38 -> 27.22 | Improved |
| use clustering regularizer instead of diffuse-specular correlation (ablation) | PBIR pipeline · Stanford-ORB relighting | PSNR-H: 26.37 -> 27.22 | Improved |
| optimize after diffusion initialization instead of direct back-projection (ablation) | Diffusion-BP · Stanford-ORB relighting | PSNR-H: 26.06 -> 26.11 | No clear effect |
| use concatenated G-buffer kernel instead of split-channel kernels (ablation) | clustering regularizer · Synthetic4Relight (4 scenes) | roughness MSE: 0.024 -> 0.013 | Improved |
| use concatenated kernel instead of split-channel (ablation) | clustering regularizer · Synthetic4Relight (4 scenes) | relighting PSNR: 32.18 -> 32.26 | No clear effect |
| add regularizer to RGB<->X upstream model (ablation) | RGB<->X guidance · Stanford-ORB relighting | PSNR-L: 32.51 -> 34.25 | Improved |
| replace IRGS smoothness terms with our regularization (ablation) | IRGS · Synthetic4Relight (4 scenes avg) | relighting PSNR: 34.78 -> 35.58 | Improved |
| use video mode instead of image mode for Diffusion-BP (ablation) | Diffusion-BP · Stanford-ORB relighting | PSNR-L: 32.67 -> 33.09 | Improved |
Closest papers by meaning
- Inverse Rendering using Multi-Bounce Path Tracing and Reservoir Sampling (ICLR 2025 · similarity 0.90)
- Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models (CVPR 2025 · similarity 0.89)
- Multi-times Monte Carlo Rendering for Inter-reflection Reconstruction (NeurIPS 2024 · similarity 0.89)
- Eclipse: Disambiguating Illumination and Materials using Unintended Shadows (CVPR 2024 · similarity 0.89)
- PBR-NeRF: Inverse Rendering with Physics-Based Neural Fields (CVPR 2025 · similarity 0.88)
- Material Palette: Extraction of Materials from a Single Image (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
